error: DWT implements no cycle counter. Cannot be used for cycle counting.

[moose]

Hello, I tried to implement the zephyr executing time functions API on my nrf52 hardware and I got the above error. According to my SoC documentation (nRF52833), this hardware does include a DWT counter. I am not sure if I need to do anything special to configure this hardware or if I need to use a relevant nrfx driver. My code was very similar to the example code in the zephyr documentation mentioned here https://docs.zephyrproject.org/latest/reference/timing_functions/index.html. Any inputs? Thx!

[moose]

Any ideas?

[Sigurd]

Hi,

Try setting this in prj.conf:

CONFIG_TIMING_FUNCTIONS=y

[moose]

I already have this config statement in my proj.conf.

[Sigurd]

Hi,

Here is my project, I used NCS v1.5.0, and I'm not able to re-create this when building for board nrf52833dk_nrf52833

timing.zip

If you are not using nrf52833dk_nrf52833 , then it might be an issue with your board files.

[davege]

Hello,

I'm experiencing exactly the same problem using nrf52840 DK (PCA10056).

If I set CONFIG_TIMING_FUNCTIONS=y, by default the Timing Zephyr subsystem will pick the architecture-dependent implementation (arch_).

So, when invoking timing_init, this function in zephyr\arch\arm\include\aarch32\cortex_m\dwt.h will be invoked:

/**
 * @brief Initialize and Enable the DWT cycle counter
 *
 * This routine enables the cycle counter and initializes its value to zero.
 *
 * @return 0
 */
static inline int z_arm_dwt_init_cycle_counter(void)
{
	/* Clear and enable the cycle counter */
	DWT->CYCCNT = 0;
	DWT->CTRL |= DWT_CTRL_CYCCNTENA_Msk;

	/* Assert that the cycle counter is indeed implemented. */
	__ASSERT((DWT->CTRL & DWT_CTRL_NOCYCCNT_Msk) != 0,
		"DWT implements no cycle counter. "
		"Cannot be used for cycle counting\n");

	return 0;
}

And contextually, the assert is triggered.

I think this is easily reproducible, I'm using a standard dev kit atm.
I don't have any other board to use with the same chip, but I don't see any reason why it would mean that this board is faulty...

Can anybody try to confirm this?
It maybe a timing issue in the asserting condition.

Note: I initially wondered why a benchmark sample (latency measure) was working fine measuring time with DWT. I then realised that asserts are just disabled in the sample...

...
CONFIG_FORCE_NO_ASSERT=y
...

Thanks in advance for your help!

D